眼底动脉硬化风险预测模型的构建和验证研究

目的]开发和验证一个眼底动脉硬化风险预测模型。 [方法]选择2006—2013年在华东疗养院体检的非眼底动脉硬化者作为开发队列,采用Lasso方法筛选预测因子,Cox回归方法建立预测模型。模型以在线计算器呈现,内部验证采用Bootstrap法完成,选取2015—2021年在该医院的体检者进行时段验证。采用一致性(C)统计量来量化区分度,并通过校准图以及Kaplan-Meier方法比较预测的生存概率与观察到的生存概率的差异来评估模型的校准度。 [结果]开发和外部验证队列分别包括33 218例和53 863例体检者。Lasso回归交叉验证结果显示,系数不为0的变量有9个。最终的Cox回归模型包括年龄、体质指数、饮酒情况、舒张压、高血压、空腹血糖、糖尿病、甘油三酯和血清尿酸9个预测因子。在线计算器网址为:https://rui2022.shinyapps.io/DynNomapp/。内部验证的C统计量为0.841,外部验证的C统计量为0.856。在开发和外部验证队列中,校准度均表现良好。 [结论]建立的眼底动脉硬化风险预测模型在体检人群中有较好的预测能力。该模型只需医院常规获取的变量,因此可以应用于体检人群的个体化管理以及高风险者管理的决策支持。

Development and validation of a risk prediction model for fundus arteriosclerosis

Aim To develop and validate a risk prediction model for fundus arteriosclerosis. Methods Patients without fundus arteriosclerosis who underwent physical examination in Huadong Sanatorium from 2006 to 2013 were selected as the derivation cohort. Lasso method was used to screen the predictors, and Cox regression method was used to establish the prediction model. The model was presented as an online calculator. The Bootstrap method was used for internal validation, and the physical examination subjects in the hospital from 2015 to 2021 were selected for temporal validation. The concordance (C) statistic was used to quantify discrimination, and the calibration of the model was evaluated by comparing the predicted survival probability with the observed survival probability using calibration plots and the Kaplan-Meier method. Results The derivation cohort included 33 218 participants and external validation cohort included 53 863 participants. The final model included nine predictors:age, body mass index, alcohol consumption, diastolic blood pressure, hypertension, fasting blood glucose, diabetes, triglyceride, and serum uric acid. Online calculator site at:https://rui2022.shinyapps.io/DynNomapp/. The C statistic of internal validation was 0.841, and the C statistic of external validation was 0.856. Calibration performed well in both the derivation and external validation cohorts. Conclusion The established risk prediction model of fundus arteriosclerosis has a good predictive ability in the physical examination population. The model only needs the variables routinely obtained by the hospital, so it can be applied to the individualized management of physical examination population and the decision support of the management of high-risk people.